1 00:00:05,720 --> 00:00:06,160 Speaker 1: Kiotra. 2 00:00:06,280 --> 00:00:09,320 Speaker 2: I'm Chelsea Daniels and this is the Front Page, a 3 00:00:09,400 --> 00:00:19,120 Speaker 2: daily podcast presented by the New Zealand Herald. Artificial intelligence 4 00:00:19,640 --> 00:00:23,000 Speaker 2: once something you'd only find in sci fi novels, it's 5 00:00:23,040 --> 00:00:24,800 Speaker 2: now an everyday. 6 00:00:24,320 --> 00:00:25,480 Speaker 1: Necessity for some. 7 00:00:26,400 --> 00:00:30,840 Speaker 2: A new accenture report for Microsoft forecasts KIWI workers are 8 00:00:30,880 --> 00:00:33,720 Speaker 2: set to save an average of two hundred and seventy 9 00:00:33,840 --> 00:00:38,760 Speaker 2: five hours a year through generative AI adoption. Today on 10 00:00:38,800 --> 00:00:43,120 Speaker 2: the front Page, Otago University's Center of Artificial Intelligence and 11 00:00:43,159 --> 00:00:47,560 Speaker 2: Public Policy Director James McLaurin is with us to discuss 12 00:00:47,680 --> 00:00:54,440 Speaker 2: the tech that's on everyone's lips. James, can you start 13 00:00:54,480 --> 00:00:59,160 Speaker 2: by explaining what AI is in the most simplest terms 14 00:00:59,160 --> 00:00:59,680 Speaker 2: if you can? 15 00:01:00,360 --> 00:01:04,480 Speaker 3: Okay, there's not really an agreed upon standard definition, but 16 00:01:04,520 --> 00:01:08,680 Speaker 3: there are simple ones. A good one is artificial intelligence 17 00:01:08,920 --> 00:01:12,520 Speaker 3: is a system that does things that people do by thinking. 18 00:01:12,840 --> 00:01:15,240 Speaker 4: That's a really old one from a go called Marvin Minsky. 19 00:01:15,560 --> 00:01:17,080 Speaker 4: It's not much use. 20 00:01:17,280 --> 00:01:19,200 Speaker 3: I mean, after all, in a sense of self opening 21 00:01:19,240 --> 00:01:22,200 Speaker 3: door is doing that. And the reason there's not agreement 22 00:01:22,360 --> 00:01:25,560 Speaker 3: on this is that really AI is a whole bunch 23 00:01:25,760 --> 00:01:29,160 Speaker 3: of different technologies that we bundle together. 24 00:01:29,040 --> 00:01:32,120 Speaker 4: Because it is taking some cognitive work away from you 25 00:01:32,160 --> 00:01:32,479 Speaker 4: and I. 26 00:01:32,959 --> 00:01:36,200 Speaker 2: It feels like it's a very hot new kind of topic, right, 27 00:01:36,319 --> 00:01:39,880 Speaker 2: But really we've been using forms of AI for years, 28 00:01:39,959 --> 00:01:40,399 Speaker 2: haven't we. 29 00:01:40,760 --> 00:01:43,440 Speaker 3: Yeah, we have, again, depending a bit on how you 30 00:01:43,520 --> 00:01:45,640 Speaker 3: define it, but a good way of thinking about simple 31 00:01:45,640 --> 00:01:48,880 Speaker 3: AI is it's built to do a particular job. New 32 00:01:48,960 --> 00:01:52,880 Speaker 3: Zealand's been using that for ages, and actually people are 33 00:01:52,960 --> 00:01:55,200 Speaker 3: very used to that sort of AI. You know, when 34 00:01:55,240 --> 00:01:57,120 Speaker 3: you type in a question to Google and it gives 35 00:01:57,120 --> 00:01:59,920 Speaker 3: you a page of links. That's an I I called pagering. 36 00:02:00,200 --> 00:02:02,480 Speaker 3: That's you know, just doing that task. It's good at 37 00:02:02,480 --> 00:02:05,559 Speaker 3: searching the internet, not for anything else. The thing that's 38 00:02:05,640 --> 00:02:08,560 Speaker 3: got really exciting in the public's mind in the last 39 00:02:08,600 --> 00:02:11,280 Speaker 3: couple of years, I guess for experts maybe over the 40 00:02:11,360 --> 00:02:14,320 Speaker 3: last four or five years, is that now we have 41 00:02:14,400 --> 00:02:17,760 Speaker 3: an that's much much more general. You know, we've always 42 00:02:17,880 --> 00:02:21,800 Speaker 3: had general AI like Siri, but it wasn't really very good, 43 00:02:22,240 --> 00:02:24,639 Speaker 3: wasn't good at answering questions, didn't know much, made lots 44 00:02:24,639 --> 00:02:27,120 Speaker 3: of mistakes. Now, all of a sudden, we have things 45 00:02:27,160 --> 00:02:31,919 Speaker 3: like chat GPT that are really good at solving basic 46 00:02:31,960 --> 00:02:35,680 Speaker 3: problems for us, interpreting what we say, giving us the 47 00:02:35,800 --> 00:02:37,720 Speaker 3: sort of outputs that we want, the sort of. 48 00:02:37,720 --> 00:02:40,360 Speaker 4: Things that we can work with. Suddenly it's become a 49 00:02:40,480 --> 00:02:41,119 Speaker 4: useful thing. 50 00:02:41,440 --> 00:02:44,360 Speaker 2: Do you think people are concerned with A I say, 51 00:02:44,480 --> 00:02:47,240 Speaker 2: like they were when microwaves or cell phones were invented. 52 00:02:47,280 --> 00:02:50,360 Speaker 2: It's new, too convenient. Perhaps there has to be something 53 00:02:50,400 --> 00:02:51,000 Speaker 2: wrong with it. 54 00:02:51,240 --> 00:02:55,440 Speaker 3: I agree people have concerns about big new things. It's 55 00:02:55,520 --> 00:02:58,600 Speaker 3: not surprising if I find it difficult to explain what 56 00:02:58,639 --> 00:03:01,760 Speaker 3: AI is and my field. Then for lots of people 57 00:03:01,840 --> 00:03:03,840 Speaker 3: it's a bit of a closed book. They don't know 58 00:03:03,919 --> 00:03:06,360 Speaker 3: how it works. People seem to be using it for 59 00:03:06,440 --> 00:03:08,720 Speaker 3: all sorts of things, and of course there are plenty 60 00:03:09,240 --> 00:03:12,680 Speaker 3: of news stories out there saying, you know, it's biased, 61 00:03:12,880 --> 00:03:15,400 Speaker 3: or it gets things wrong, which it does from time 62 00:03:15,440 --> 00:03:18,760 Speaker 3: to time, or it's going to be used by bad 63 00:03:18,800 --> 00:03:21,720 Speaker 3: actors to promote disinformation or something like that. 64 00:03:21,919 --> 00:03:25,000 Speaker 4: So there are certainly lots of bad news stories. I think. 65 00:03:25,040 --> 00:03:27,800 Speaker 3: The good way I think about AI is it's so 66 00:03:28,120 --> 00:03:31,320 Speaker 3: general purpose. It's like the development of the Internet that 67 00:03:31,480 --> 00:03:33,560 Speaker 3: is going to be used for good and is going 68 00:03:33,600 --> 00:03:35,680 Speaker 3: to be used for bad. You know, That's what you 69 00:03:35,760 --> 00:03:38,360 Speaker 3: get with these really general purpose technologies. 70 00:03:38,720 --> 00:03:43,520 Speaker 2: An Accentua report has found AI could grow GDP by 71 00:03:43,640 --> 00:03:46,800 Speaker 2: one percent a year and at about seventy six billion 72 00:03:46,960 --> 00:03:49,800 Speaker 2: with a b a year to New Zealand's economy by 73 00:03:49,800 --> 00:03:50,720 Speaker 2: twenty thirty eight. 74 00:03:50,880 --> 00:03:52,160 Speaker 1: Does that sound about right to you? 75 00:03:52,440 --> 00:03:55,560 Speaker 3: Sounds low to me, But look, the real issue is 76 00:03:56,080 --> 00:03:58,520 Speaker 3: that it's early days yet. This is the first part 77 00:03:58,520 --> 00:04:01,640 Speaker 3: of a long cricket match, so thinking about this is open. 78 00:04:01,680 --> 00:04:05,880 Speaker 3: AYE has recently given us a little list of sort 79 00:04:05,880 --> 00:04:10,600 Speaker 3: of waypoints in the development of artificial intelligence, and it's 80 00:04:10,600 --> 00:04:13,640 Speaker 3: got five of these waypoints, and it thinks we're really 81 00:04:13,760 --> 00:04:15,160 Speaker 3: only at the first one. 82 00:04:15,760 --> 00:04:18,159 Speaker 4: So the first one is AI that's good. 83 00:04:17,960 --> 00:04:22,080 Speaker 3: At answering questions, so that includes chatbots like chat GPT, 84 00:04:22,480 --> 00:04:24,839 Speaker 3: but also things that can draw you a picture or 85 00:04:24,880 --> 00:04:28,640 Speaker 3: make you a video, or that can give you advice 86 00:04:28,680 --> 00:04:31,120 Speaker 3: about what to do, like something that's built into an 87 00:04:31,200 --> 00:04:36,320 Speaker 3: autonomous car. The next step is things that can reason. 88 00:04:36,480 --> 00:04:38,480 Speaker 3: And the difference between the chatbots and the reason is 89 00:04:38,480 --> 00:04:41,160 Speaker 3: is that the chatbots are sort of following rules that 90 00:04:41,200 --> 00:04:43,919 Speaker 3: are already there, and the reasons that are answering really 91 00:04:43,960 --> 00:04:46,680 Speaker 3: difficult problems, so they're having to make up their own rules. 92 00:04:46,960 --> 00:04:49,720 Speaker 3: The next step after that is agents, and you can 93 00:04:49,720 --> 00:04:52,120 Speaker 3: think of an agent as being like an assistant. You know, 94 00:04:52,200 --> 00:04:54,560 Speaker 3: I tell my assistant I want to go on holiday, 95 00:04:54,760 --> 00:04:56,920 Speaker 3: could you do it for me? Now they know some 96 00:04:56,960 --> 00:04:59,239 Speaker 3: facts about me. They know all sorts of things about 97 00:04:59,560 --> 00:05:01,360 Speaker 3: going on holiday, the types of things you. 98 00:05:01,320 --> 00:05:01,839 Speaker 4: Have to book. 99 00:05:02,440 --> 00:05:05,080 Speaker 3: They then go out, go to all the sites, use 100 00:05:05,120 --> 00:05:07,039 Speaker 3: all the tools, do all the things that are a 101 00:05:07,080 --> 00:05:09,520 Speaker 3: really good assistant will do, and then come back to 102 00:05:09,520 --> 00:05:10,840 Speaker 3: me and say, here you go. 103 00:05:10,920 --> 00:05:12,640 Speaker 4: You go into the Bay of Islands, here's where you're 104 00:05:12,640 --> 00:05:14,920 Speaker 4: gonna stay. We're gonna reach your car someone and so forth. 105 00:05:15,080 --> 00:05:18,000 Speaker 3: The next step after that, which we really don't have yet, 106 00:05:18,120 --> 00:05:21,160 Speaker 3: is the innovator that comes up with, you know, here's 107 00:05:21,200 --> 00:05:22,400 Speaker 3: how we're going to cure cancer. 108 00:05:22,720 --> 00:05:25,720 Speaker 4: There are systems that are being developed. There's something called 109 00:05:25,760 --> 00:05:29,160 Speaker 4: the AI scientists that is just the start of work 110 00:05:29,200 --> 00:05:31,240 Speaker 4: on this. And then the final. 111 00:05:30,880 --> 00:05:34,400 Speaker 3: Step says open AI is ais that could do what 112 00:05:34,520 --> 00:05:38,120 Speaker 3: a whole big organization did, like your whole city council 113 00:05:38,279 --> 00:05:39,240 Speaker 3: or a whole company. 114 00:05:39,360 --> 00:05:41,799 Speaker 4: We don't have anything like that yet. 115 00:05:42,279 --> 00:05:44,520 Speaker 2: How long do you think it'll take us to get 116 00:05:44,560 --> 00:05:45,359 Speaker 2: to each point? 117 00:05:45,720 --> 00:05:49,200 Speaker 3: Science is this frustrating thing because it's a creative process, 118 00:05:49,279 --> 00:05:52,440 Speaker 3: so it's really hard to predict. It is sensitive to 119 00:05:52,480 --> 00:05:54,040 Speaker 3: some things, and one of the things that it's very 120 00:05:54,080 --> 00:05:56,240 Speaker 3: sensitive to is lots of money. And there is a 121 00:05:56,520 --> 00:06:00,840 Speaker 3: mountain of money being poured in many, many billions of dollars, 122 00:06:00,960 --> 00:06:03,440 Speaker 3: hundreds of billions of dollars being poured into AI at 123 00:06:03,480 --> 00:06:06,039 Speaker 3: the moment. Maybe a way that you could answer this 124 00:06:06,240 --> 00:06:09,360 Speaker 3: is to say that there are surveys of computer scientists 125 00:06:09,360 --> 00:06:12,440 Speaker 3: and experts and machine learning asking them when they think 126 00:06:12,520 --> 00:06:15,719 Speaker 3: we'll get all the way to these super powerful AIS, 127 00:06:16,440 --> 00:06:19,440 Speaker 3: and it's common for people to think that that might 128 00:06:19,600 --> 00:06:20,720 Speaker 3: take ten years. 129 00:06:21,160 --> 00:06:25,440 Speaker 2: What New Zealand industries could benefit the most from adopting 130 00:06:25,520 --> 00:06:26,960 Speaker 2: AI like now? 131 00:06:27,160 --> 00:06:30,240 Speaker 3: So in the world of people who study work, they 132 00:06:30,279 --> 00:06:34,080 Speaker 3: talk about general purpose technologies. A general purpose technology is 133 00:06:34,120 --> 00:06:38,200 Speaker 3: something like electricity, or the motive vehicle, or the production line, 134 00:06:38,279 --> 00:06:41,760 Speaker 3: or you know, something that's got really open ended possible uses. 135 00:06:42,000 --> 00:06:46,720 Speaker 3: AIS arenambiguously a general purpose technology, so you know you 136 00:06:46,760 --> 00:06:50,520 Speaker 3: can use it pretty much everywhere. Maybe the best way 137 00:06:50,680 --> 00:06:52,240 Speaker 3: to answer this to say, you know, what are the 138 00:06:52,279 --> 00:06:55,920 Speaker 3: biggest problems for New Zealand. Things that are really expensive 139 00:06:55,960 --> 00:06:58,880 Speaker 3: that are big parts of our national budgets, So things 140 00:06:58,960 --> 00:07:03,040 Speaker 3: like health and education, productivity as a whole, all these 141 00:07:03,040 --> 00:07:05,640 Speaker 3: things are things that will benefit Indeed, some of them 142 00:07:05,640 --> 00:07:08,400 Speaker 3: are already benefiting from the use of AI. 143 00:07:09,200 --> 00:07:10,880 Speaker 4: There's going to be a lot of use of AI 144 00:07:11,400 --> 00:07:12,880 Speaker 4: in transport. 145 00:07:12,960 --> 00:07:16,800 Speaker 3: We already know we're pretty close to autonomous vehicles. That's 146 00:07:16,840 --> 00:07:18,760 Speaker 3: going to put it adjacent to lots and lots of 147 00:07:18,760 --> 00:07:21,239 Speaker 3: the economy because lots of things have a transport cost 148 00:07:21,360 --> 00:07:25,760 Speaker 3: built into them. Really good at sort of back office functions, 149 00:07:25,880 --> 00:07:27,520 Speaker 3: administrative functions. 150 00:07:28,000 --> 00:07:30,840 Speaker 4: One of the uses that people are putting. 151 00:07:30,520 --> 00:07:34,200 Speaker 3: AI to at the moment is as a helper for 152 00:07:34,760 --> 00:07:39,040 Speaker 3: doctors for GPS. GPS spend a lot of their time 153 00:07:39,440 --> 00:07:43,080 Speaker 3: taking notes, So one thing AI is really good at 154 00:07:43,240 --> 00:07:46,160 Speaker 3: is listening to conversations and writing out what happened in 155 00:07:46,200 --> 00:07:48,360 Speaker 3: the conversation in a way that you tell it to 156 00:07:48,400 --> 00:07:50,400 Speaker 3: write it up, and it gives it to you at 157 00:07:50,400 --> 00:07:52,600 Speaker 3: the end, and then you just go through it and say, yes, 158 00:07:52,640 --> 00:07:54,600 Speaker 3: I like that, I want to edit that. But things 159 00:07:54,680 --> 00:07:58,840 Speaker 3: like that, you know, a huge help in domains where 160 00:07:58,880 --> 00:08:02,080 Speaker 3: we just don't have an people at the moment, and 161 00:08:02,800 --> 00:08:05,280 Speaker 3: you know, we're pressed for people to get medical appointments, 162 00:08:05,360 --> 00:08:06,000 Speaker 3: things like that. 163 00:08:06,280 --> 00:08:09,840 Speaker 2: There's been some backlash in some quarters for when AI 164 00:08:09,880 --> 00:08:10,560 Speaker 2: is used. 165 00:08:11,080 --> 00:08:15,040 Speaker 5: Marvel's new TV series Secret in Vision has a controversial intro. 166 00:08:15,960 --> 00:08:19,480 Speaker 5: The animated intro was done using AI, which is angering 167 00:08:19,560 --> 00:08:22,880 Speaker 5: fans and creators. The backlash against the use of AI 168 00:08:23,280 --> 00:08:27,160 Speaker 5: traces back in parts to concerns about studios replacing creative 169 00:08:27,160 --> 00:08:28,239 Speaker 5: workers with AI. 170 00:08:28,920 --> 00:08:32,400 Speaker 2: Do organizations need to be crystal clear when they're using 171 00:08:32,480 --> 00:08:35,199 Speaker 2: AI and is there perhaps a need to be mindful 172 00:08:35,280 --> 00:08:37,920 Speaker 2: of using this technology too much? I guess in the 173 00:08:37,960 --> 00:08:40,240 Speaker 2: place of human creativity, I. 174 00:08:40,200 --> 00:08:44,760 Speaker 3: Think it certainly helped to be very open about when 175 00:08:44,800 --> 00:08:48,040 Speaker 3: you're using it, if only because you know, if something 176 00:08:48,080 --> 00:08:51,200 Speaker 3: goes wrong and you haven't told the public or your client, 177 00:08:51,600 --> 00:08:54,560 Speaker 3: you know that you're using this tool, then you know 178 00:08:54,679 --> 00:08:57,040 Speaker 3: that looks bad for you, that's going to hurt your reputation. 179 00:08:57,720 --> 00:09:01,840 Speaker 3: So I think that sort of clarity really important. I'm 180 00:09:01,880 --> 00:09:05,480 Speaker 3: pretty sure that there are domains where people won't like 181 00:09:05,720 --> 00:09:08,439 Speaker 3: the use of it, you know, where people will prefer 182 00:09:08,600 --> 00:09:10,640 Speaker 3: to deal with a human or want to know that 183 00:09:10,679 --> 00:09:13,720 Speaker 3: it's a human, for example, creating this artwork rather than 184 00:09:13,720 --> 00:09:14,240 Speaker 3: a machine. 185 00:09:14,360 --> 00:09:15,160 Speaker 4: Something like that. 186 00:09:15,160 --> 00:09:19,680 Speaker 3: That makes sense people are going to rebel against its 187 00:09:19,800 --> 00:09:23,200 Speaker 3: use in general. I don't think so, because it's so 188 00:09:23,520 --> 00:09:25,920 Speaker 3: general purpose that it would be like rebelling against the 189 00:09:26,000 --> 00:09:26,840 Speaker 3: use of the Internet. 190 00:09:27,200 --> 00:09:28,440 Speaker 4: But I keep saying it's. 191 00:09:28,360 --> 00:09:32,000 Speaker 3: Early days yet, and as it is more disruptive in 192 00:09:32,160 --> 00:09:36,880 Speaker 3: more hearts of the economy, we'll certainly see places where 193 00:09:37,040 --> 00:09:40,320 Speaker 3: people will argue against its use. So in Hollywood, as 194 00:09:40,760 --> 00:09:44,640 Speaker 3: you referred to before, screen writers have argued that it 195 00:09:44,720 --> 00:09:48,760 Speaker 3: mustn't be used because it's taken away creative jobs and 196 00:09:49,000 --> 00:09:52,280 Speaker 3: people want to have people, not machines writing their. 197 00:09:52,200 --> 00:09:53,200 Speaker 4: Scripts for them. 198 00:09:53,559 --> 00:09:56,680 Speaker 3: Legally, the end of that dispute has been that it 199 00:09:56,760 --> 00:09:59,560 Speaker 3: is okay to use AI in this context, but you 200 00:09:59,640 --> 00:10:01,160 Speaker 3: have to use humans as well. 201 00:10:11,360 --> 00:10:15,199 Speaker 2: Technology Minister Judith Collins presented a paper to Cabinet called 202 00:10:15,280 --> 00:10:19,160 Speaker 2: Approach to Work on Artificial Intelligence that was in June. 203 00:10:19,480 --> 00:10:21,760 Speaker 2: I'm sure you've read it cover to cover, but in it, 204 00:10:22,080 --> 00:10:26,400 Speaker 2: she said, New Zealanders are often early adopters of new technology, 205 00:10:26,400 --> 00:10:29,480 Speaker 2: but businesses are slow to adopt AI due in part 206 00:10:29,520 --> 00:10:33,280 Speaker 2: to uncertainty about the future regulatory environment. 207 00:10:33,320 --> 00:10:35,400 Speaker 1: Would you agree with that, Yes, that. 208 00:10:35,440 --> 00:10:38,080 Speaker 3: New Zealand is pretty good at being an early adopter. 209 00:10:38,320 --> 00:10:41,880 Speaker 3: I think that's right. Wherever you get a transfer of technologies, 210 00:10:41,920 --> 00:10:45,040 Speaker 3: it's a very fraught thing for businesses. Nice example of 211 00:10:45,080 --> 00:10:48,080 Speaker 3: that at the moment is the move from internal combustion 212 00:10:48,200 --> 00:10:51,880 Speaker 3: cars to electric cars. So, after all, making electric car 213 00:10:52,040 --> 00:10:55,120 Speaker 3: isn't really very like making an internal combustion engine car. 214 00:10:55,200 --> 00:10:58,120 Speaker 3: It's superficially like, but not under the hood. So in 215 00:10:58,240 --> 00:11:02,319 Speaker 3: order to get industry is to you know, take the leap. 216 00:11:02,559 --> 00:11:04,760 Speaker 3: They need a lot of information, they need a lot 217 00:11:04,800 --> 00:11:07,360 Speaker 3: of support, They need to do quite a lot of 218 00:11:07,400 --> 00:11:10,240 Speaker 3: work to think about what their industry looks like and 219 00:11:10,280 --> 00:11:12,520 Speaker 3: what they look like as a company. And if they 220 00:11:12,520 --> 00:11:14,840 Speaker 3: don't jump, is one of their competitors going to jump? 221 00:11:15,040 --> 00:11:17,720 Speaker 3: And how could they do this in a way that's 222 00:11:17,760 --> 00:11:20,720 Speaker 3: going to benefit them as a company and benefit their reputation. 223 00:11:21,000 --> 00:11:22,880 Speaker 3: But we do this, you know, we have these big 224 00:11:22,920 --> 00:11:26,320 Speaker 3: technology changes every few decades, and I no doubt news 225 00:11:26,360 --> 00:11:29,320 Speaker 3: in and will achieve it. I do think that it's 226 00:11:29,360 --> 00:11:31,560 Speaker 3: important for businesses to be proactive. 227 00:11:31,800 --> 00:11:34,680 Speaker 2: It kind of reminds me of quotes from people who 228 00:11:35,160 --> 00:11:38,440 Speaker 2: missed out, perhaps on buying shares in Google or Apple 229 00:11:38,520 --> 00:11:42,319 Speaker 2: or Microsoft. Right, is that the same case with AI 230 00:11:42,679 --> 00:11:45,400 Speaker 2: or is it okay to take a step back and 231 00:11:45,520 --> 00:11:49,960 Speaker 2: make sure everything is above board before really diving in? 232 00:11:50,200 --> 00:11:53,080 Speaker 3: Yep, you absolutely want to do your due diligence before 233 00:11:53,120 --> 00:11:56,240 Speaker 3: you dive in. You know, everybody's using AI. This is 234 00:11:56,240 --> 00:11:59,600 Speaker 3: some MAI, so let's use this. That's a very risky strategy. 235 00:12:00,160 --> 00:12:02,319 Speaker 3: It might be that at the end of your due diligence, 236 00:12:02,559 --> 00:12:05,920 Speaker 3: it's just not the thing that lots of companies are doing, 237 00:12:06,000 --> 00:12:08,440 Speaker 3: doesn't work for you, and you make a judgment call 238 00:12:08,520 --> 00:12:11,440 Speaker 3: that the tools aren't yet available or it's just not 239 00:12:11,520 --> 00:12:13,720 Speaker 3: something you want to do. I'm not saying everybody's got 240 00:12:13,760 --> 00:12:16,280 Speaker 3: to get out there and use it, but really everybody 241 00:12:16,320 --> 00:12:20,120 Speaker 3: should be having a careful think about it, evaluating it, 242 00:12:20,240 --> 00:12:24,480 Speaker 3: trying to find some expertise, get out there, do your 243 00:12:24,520 --> 00:12:27,760 Speaker 3: own research, get yourself used to what it is and 244 00:12:27,800 --> 00:12:28,480 Speaker 3: how it works. 245 00:12:28,920 --> 00:12:32,400 Speaker 2: In terms of the regulatory framework, does that need to 246 00:12:32,440 --> 00:12:33,920 Speaker 2: be set up right now? 247 00:12:34,200 --> 00:12:39,600 Speaker 3: New Zealand has benefited in the past from regulations that. 248 00:12:39,520 --> 00:12:42,520 Speaker 4: Were pasted elsewhere. You know, we're a little country. 249 00:12:42,800 --> 00:12:45,720 Speaker 3: We don't have enough clout to be pushing around the 250 00:12:45,760 --> 00:12:46,559 Speaker 3: Googles and the. 251 00:12:46,520 --> 00:12:47,760 Speaker 4: Microsofts of this world. 252 00:12:48,240 --> 00:12:53,080 Speaker 3: But there are places that regulate much more quickly. In general, 253 00:12:53,120 --> 00:12:56,080 Speaker 3: regulate pretty well. The EU is a good example of that. 254 00:12:56,440 --> 00:12:59,439 Speaker 3: So in the last wave of AI, the EU passed 255 00:12:59,480 --> 00:13:03,120 Speaker 3: something called the GENPR General Data Protection. 256 00:13:02,840 --> 00:13:05,640 Speaker 4: Rules that was pretty useful. 257 00:13:06,440 --> 00:13:09,599 Speaker 3: We didn't pass mirroring legislation, I guess we could have, 258 00:13:09,720 --> 00:13:13,200 Speaker 3: but New Zenader's got some benefit from that. The EU 259 00:13:13,320 --> 00:13:16,800 Speaker 3: has just passed something called the AI Act. We will 260 00:13:16,880 --> 00:13:19,320 Speaker 3: get some benefit from that. 261 00:13:21,160 --> 00:13:24,040 Speaker 6: There are certain things which will be prohibited in the 262 00:13:24,080 --> 00:13:26,720 Speaker 6: European Union. One of these is, for example, that you 263 00:13:26,760 --> 00:13:30,080 Speaker 6: will not be allowed to do a facial recognition on 264 00:13:30,200 --> 00:13:34,920 Speaker 6: CCTV live streams except for the authorities in very very 265 00:13:35,200 --> 00:13:37,400 Speaker 6: restricted circumstances. 266 00:13:39,360 --> 00:13:43,160 Speaker 3: Because this is such an expensive sort of technology to build, 267 00:13:43,520 --> 00:13:47,320 Speaker 3: New Zenadors mostly are going to be importing this rather 268 00:13:47,360 --> 00:13:48,800 Speaker 3: than building it here. 269 00:13:49,120 --> 00:13:50,319 Speaker 4: You know, we import cars. 270 00:13:50,320 --> 00:13:52,040 Speaker 3: We don't really make cars in New Zenand, but we 271 00:13:52,080 --> 00:13:54,160 Speaker 3: import them and we get a huge amount of economic 272 00:13:54,240 --> 00:13:55,480 Speaker 3: value out of having them. 273 00:13:55,679 --> 00:13:57,360 Speaker 4: So the task for us is. 274 00:13:57,280 --> 00:13:59,680 Speaker 3: To work out, right, how do we use the cars, 275 00:13:59,720 --> 00:14:02,679 Speaker 3: what rules we want to have around you know, the 276 00:14:02,720 --> 00:14:05,560 Speaker 3: cars that you can buy and how you'll be licensed 277 00:14:05,559 --> 00:14:07,240 Speaker 3: to use them, and how often we want to check 278 00:14:07,320 --> 00:14:09,600 Speaker 3: to see whether they're own worthy. So those kind of 279 00:14:09,760 --> 00:14:13,520 Speaker 3: audit tasks I think are going to become very important 280 00:14:13,640 --> 00:14:18,400 Speaker 3: for New Zealand. Whether in the end we pass regulations, 281 00:14:19,080 --> 00:14:22,120 Speaker 3: I don't know. New Zealand's had sort of voluntary frameworks, 282 00:14:22,160 --> 00:14:24,760 Speaker 3: the Principles for Sake and Effective Use of Data and Analytics, 283 00:14:24,760 --> 00:14:28,000 Speaker 3: the Algorithmic Charter, things like that in the past for 284 00:14:28,200 --> 00:14:31,920 Speaker 3: government use of AI, and they've worked pretty well. I 285 00:14:32,000 --> 00:14:35,880 Speaker 3: think it's early days yet to work out whether or not, 286 00:14:36,480 --> 00:14:39,480 Speaker 3: you know, we want to update general sets of rules 287 00:14:39,600 --> 00:14:44,520 Speaker 3: like this, or encourage the development of regulations, or just 288 00:14:44,560 --> 00:14:47,840 Speaker 3: in a sensible business rules in particular domains. I've been 289 00:14:47,840 --> 00:14:49,360 Speaker 3: doing quite a bit of work with the Ministry of 290 00:14:49,400 --> 00:14:52,800 Speaker 3: Health lately thinking about the use of AI in healthcare 291 00:14:52,840 --> 00:14:56,360 Speaker 3: in New Zealand, and that feels like a good sort 292 00:14:56,360 --> 00:14:57,840 Speaker 3: of level of granularity. 293 00:14:58,200 --> 00:14:59,560 Speaker 4: So it might be a sort of thing that we 294 00:14:59,600 --> 00:15:02,120 Speaker 4: go int by industry or demain by domain. 295 00:15:02,600 --> 00:15:05,360 Speaker 2: Will we be the all losing our jobs due to AI? 296 00:15:05,720 --> 00:15:06,880 Speaker 2: Or is that a bit overblown? 297 00:15:06,920 --> 00:15:11,200 Speaker 3: Do you think automation doesn't very often take whole jobs? 298 00:15:11,520 --> 00:15:16,040 Speaker 3: Automation usually takes tasks out of jobs, and the same 299 00:15:16,200 --> 00:15:19,880 Speaker 3: is true for AI. I guess the thing we should 300 00:15:19,920 --> 00:15:23,960 Speaker 3: say first is tools like I do two things. They 301 00:15:24,000 --> 00:15:29,120 Speaker 3: both enhance individuals and sometimes they replace individuals. So think 302 00:15:29,160 --> 00:15:32,360 Speaker 3: of a technology like a laptop that enhances me, makes 303 00:15:32,360 --> 00:15:36,800 Speaker 3: me more valuable, more productive, whereas the baggage handling robot 304 00:15:36,840 --> 00:15:39,960 Speaker 3: at the airport is replacing somebody, it's not making somebody 305 00:15:39,960 --> 00:15:43,960 Speaker 3: more productive, it's just replacing them. So AI is going 306 00:15:44,040 --> 00:15:48,520 Speaker 3: to take some tasks out of jobs. That's going to 307 00:15:48,600 --> 00:15:52,080 Speaker 3: be fairly unpredictable for lots of people. It will be 308 00:15:52,200 --> 00:15:55,760 Speaker 3: helpful and useful, and there have been plenty of studies 309 00:15:55,800 --> 00:15:58,960 Speaker 3: done in the last year or two just of people 310 00:15:59,040 --> 00:16:02,000 Speaker 3: being given access to chet GPT in their ordinary sort 311 00:16:02,040 --> 00:16:05,040 Speaker 3: of daily workflow and told you can use this. 312 00:16:05,440 --> 00:16:06,960 Speaker 4: You know, here's a bunch of tasks, try and do 313 00:16:07,040 --> 00:16:07,680 Speaker 4: these tasks. 314 00:16:07,880 --> 00:16:10,520 Speaker 3: There's pretty good evidence that it makes people more productive 315 00:16:10,760 --> 00:16:14,080 Speaker 3: in general, that people quite like using it, that people 316 00:16:14,120 --> 00:16:16,920 Speaker 3: feel that it's doing things that are much like doing. 317 00:16:17,200 --> 00:16:20,600 Speaker 3: People report being a little nervous about, you know, the 318 00:16:20,600 --> 00:16:23,840 Speaker 3: coming of AI in context like this, but. 319 00:16:23,720 --> 00:16:26,320 Speaker 4: When asked did you like using it, most people say yes. 320 00:16:26,600 --> 00:16:29,920 Speaker 3: That's not to say that there won't be jobs where 321 00:16:30,280 --> 00:16:35,560 Speaker 3: AI changes the job in a pretty substantial way. We 322 00:16:35,760 --> 00:16:39,640 Speaker 3: don't really have AI like this yet, but imagine that 323 00:16:39,960 --> 00:16:44,120 Speaker 3: an AI comes along that can take the diagnostic task 324 00:16:44,360 --> 00:16:47,800 Speaker 3: off your family doctor. Well, that's a very important, high 325 00:16:47,880 --> 00:16:50,600 Speaker 3: value task that they do, So that would be a 326 00:16:50,640 --> 00:16:53,280 Speaker 3: sort of negative change to their job. I'm thinking they 327 00:16:53,280 --> 00:16:55,120 Speaker 3: would think it was a negative change to their job. 328 00:16:55,320 --> 00:16:58,440 Speaker 3: On the other hand, think of AI that does legal discovery. 329 00:16:58,520 --> 00:16:59,200 Speaker 4: For a lawyer. 330 00:16:59,360 --> 00:17:01,960 Speaker 3: Legal discovery is just you know, the needle in the haystack, 331 00:17:02,040 --> 00:17:03,440 Speaker 3: hunting for facts pretty trial. 332 00:17:03,520 --> 00:17:05,880 Speaker 4: This sort of thing. Well, that's drudgery, that's boring work. 333 00:17:06,080 --> 00:17:09,400 Speaker 3: So that sort of AI is making the lawyer happier, 334 00:17:09,560 --> 00:17:13,840 Speaker 3: more productive, more successful. We're going to wait and see 335 00:17:13,880 --> 00:17:16,400 Speaker 3: with the chips fall. I keep saying it's early days, 336 00:17:16,400 --> 00:17:19,000 Speaker 3: but it really is. But you know, whenever people say 337 00:17:19,160 --> 00:17:21,359 Speaker 3: is it going to take all the jobs, think about 338 00:17:21,520 --> 00:17:25,000 Speaker 3: jobs that aren't being done that you would like to 339 00:17:25,480 --> 00:17:29,080 Speaker 3: have done. You know people who are older, living in 340 00:17:29,119 --> 00:17:31,080 Speaker 3: their own homes and worried they're not going to be 341 00:17:31,080 --> 00:17:31,480 Speaker 3: able to. 342 00:17:31,400 --> 00:17:32,600 Speaker 4: Cope in their own homes. 343 00:17:33,040 --> 00:17:36,000 Speaker 3: Well, we can't have somebody in their home twenty four 344 00:17:36,000 --> 00:17:38,600 Speaker 3: hours a day to help them. We might be able 345 00:17:38,600 --> 00:17:40,920 Speaker 3: to have AI in their home twenty four hours a 346 00:17:41,000 --> 00:17:43,000 Speaker 3: day to help them. So, you know, this might be 347 00:17:43,000 --> 00:17:45,560 Speaker 3: an invention of a new task that's really valuable to 348 00:17:45,560 --> 00:17:46,040 Speaker 3: some people. 349 00:17:46,320 --> 00:17:48,239 Speaker 2: And like you said, it's early days, it's not all 350 00:17:48,240 --> 00:17:49,080 Speaker 2: happening tomorrow. 351 00:17:49,160 --> 00:17:49,440 Speaker 1: Is it. 352 00:17:49,520 --> 00:17:53,119 Speaker 2: I mean, I remember headlines when someone asked an AI 353 00:17:53,280 --> 00:17:55,879 Speaker 2: program to tell her how many ours are in the 354 00:17:55,920 --> 00:17:58,600 Speaker 2: word strawberry, and it insisted there were only two. And 355 00:17:58,600 --> 00:18:01,920 Speaker 2: we've also seen all those AI generated images where someone's 356 00:18:02,000 --> 00:18:04,439 Speaker 2: got seven fingers and maybe one leg is twice as 357 00:18:04,480 --> 00:18:05,080 Speaker 2: long as the other. 358 00:18:05,400 --> 00:18:07,160 Speaker 1: It's not close to being at any point. 359 00:18:06,960 --> 00:18:11,040 Speaker 2: Where it can comfortably replace us without people realizing it. 360 00:18:11,040 --> 00:18:12,160 Speaker 1: It's not gonna happen tomorrow. 361 00:18:12,440 --> 00:18:15,880 Speaker 3: There's an idea that's been promoted by an American academic 362 00:18:15,920 --> 00:18:18,760 Speaker 3: called Ethan Moolloch, which is called the jagged edge. 363 00:18:19,080 --> 00:18:21,280 Speaker 4: So if you think of AI as sort of. 364 00:18:21,280 --> 00:18:26,080 Speaker 3: Gradually expanding out into domains that people work, and the 365 00:18:26,160 --> 00:18:29,520 Speaker 3: jagged edge is this idea that it's not a smooth 366 00:18:29,600 --> 00:18:33,880 Speaker 3: replacement of a job. These ais are bizarrely good at 367 00:18:33,920 --> 00:18:37,520 Speaker 3: some things and bizarrely bad at other things in ways 368 00:18:37,560 --> 00:18:40,400 Speaker 3: that it's a bit hard for human beings to. 369 00:18:40,359 --> 00:18:41,439 Speaker 4: Wrap their head around. 370 00:18:42,000 --> 00:18:45,240 Speaker 3: You rightly point out that maths is a real challenge 371 00:18:45,480 --> 00:18:47,240 Speaker 3: for modern AIS. 372 00:18:47,600 --> 00:18:50,480 Speaker 4: Generous of ais as they call them. It's not the 373 00:18:50,600 --> 00:18:52,800 Speaker 4: end of the world. That is, after all a challenge 374 00:18:52,800 --> 00:18:55,080 Speaker 4: for lots of us. But we just use calculators. 375 00:18:55,640 --> 00:18:59,040 Speaker 3: But on the other hand, the same AIS can pass 376 00:18:59,280 --> 00:19:04,040 Speaker 3: chunks of of bar exams or medical practice tests much 377 00:19:04,040 --> 00:19:06,639 Speaker 3: better than the average lawyer or doctor. You know, is 378 00:19:06,640 --> 00:19:08,960 Speaker 3: it going to replace people? Is it going to come 379 00:19:09,359 --> 00:19:12,520 Speaker 3: tomorrow in some domains. Yes, we shouldn't be too complacent, 380 00:19:13,080 --> 00:19:15,040 Speaker 3: not in all domains, and we've got a lot of 381 00:19:15,040 --> 00:19:17,720 Speaker 3: work to do to think about whether this domain or 382 00:19:17,720 --> 00:19:20,399 Speaker 3: that domain is somewhere that's going to be affected quickly. 383 00:19:20,920 --> 00:19:22,280 Speaker 1: Thanks for joining us, James. 384 00:19:27,600 --> 00:19:30,399 Speaker 2: That's it for this episode of the Front Page. You 385 00:19:30,440 --> 00:19:34,280 Speaker 2: can read more about today's stories and extensive news coverage. 386 00:19:33,920 --> 00:19:36,040 Speaker 1: At NZ Herald dot co dot nz. 387 00:19:36,880 --> 00:19:39,800 Speaker 2: The Front Page is produced by Ethan Sells with sound 388 00:19:39,840 --> 00:19:41,119 Speaker 2: engineer Patty Fox. 389 00:19:41,520 --> 00:19:42,919 Speaker 1: I'm Chelsea Daniels. 390 00:19:43,480 --> 00:19:46,439 Speaker 2: Subscribe to The Front Page on iHeartRadio or wherever you 391 00:19:46,440 --> 00:19:49,840 Speaker 2: get your podcasts, and tune in tomorrow for another look 392 00:19:49,920 --> 00:19:51,120 Speaker 2: behind the headlines.